browser-to-api

Browser to API turns a website's observable HTTP traffic into a best-effort OpenAPI 3.1 specification by post-processing a capture produced by the sibling `browser-trace` skill. It solves the problem of building a client or SDK against a third-party or undocumented website that publishes no API spec: rather than reverse-engineering endpoints by hand, it reads recorded CDP request/response events and derives a documented API surface plus a coverage report. Importantly, the skill does not capture traffic itself, it is purely offline processing over an existing `.o11y/<run>/` trace directory.

The pipeline runs as ordered Node ESM stages (load, filter, normalize, infer, emit) dispatched by `discover.mjs`. It pairs CDP requests and responses by requestId, drops CORS preflights and non-API resource types, templatizes URL paths (detecting UUIDs, integers, and slugs into path variables), collects query parameters, infers JSON Schemas from observed request and response bodies (with enum, format, and required-field detection), and hoists shared schemas into components. Output includes openapi.yaml/json, a self-contained HTML report with expandable operation cards, a generated client.mjs snippet, redacted sample payloads, and a confidence report. Scripts use only the Node standard library (no npm install), include a redaction module, and can optionally preview the spec in a local Swagger UI. Response-body schemas require optionally enabling `browse network on` during capture.

It targets developers and integration engineers building clients against sites that lack published API docs, and teams who want a coverage report showing which flows would broaden a spec. It is MIT-licensed and restricted to Bash, Read, and Grep tools.